Transcription
If I was just starting Cloud Code in 2026, then these are the top 10 skills, plugins, and CLIs that I would install and start using right away. But even if you aren't brand new to Claude Code, I still think you're going to get introduced to a few you haven't seen before. And if that's the case, then I've done my job.
So number one on the list is the Codeex plugin for Claude Code. Yes, that Codeex from OpenAI. We have the ability to use Codeex in conjunction with Claude. Now, you might ask, why would I do that? Well, there's a very specific use case where you can get a ton of value out of combining these two tools, and that is CodeEx review and Codeex adversarial review. Because when we create something in Cloud Code, especially if you don't come from a technical background, it can be difficult for us to actually assess if the code it created was the best path forward. Sure, I can look at the outputs whatever Claude code gave me. But if I look at the actual code, do you know what's going on? Do I know what's going on? Not necessarily if that isn't our domain. And while I can tell cloud code, take a look at your code and review it, the truth is most large language models, opus 4.6 and sonnet 4.6 included, look very nicely on the on their own code, they're not going to come back and say, "My code sucks." However, if we use an outside agent like Codeex to do the code review for us, especially an adversarial review where it's really getting into the nuts and bolts of how we put it together and telling us this is wrong, this isn't. Well, we have a recipe for better outputs and better foundations inside of Cloud Code.
Now, installing this is very simple. If you search for Codeex plugin Cloud Code, it's going to take you to this GitHub. And adding it and then installing it to the marketplace is as simple as copying these commands, going inside of Claude Code and pasting them in there. Once you install it, you just have to run the reload plugins and then run codeex setup. To use this, you do need an open AI account. So if you have a chat GPT account, even if it's the go plan, seven bucks a month, that's more than enough. From there, you just open cloud code in whatever project you're running, and you run the command codeex colad adversarial review, and it will do all of this automatically. Now, Codex has a few other commands like a normal review that's not as hardcore. And you can also specify what exactly you want it to. It also has options like codeex rescue. So if you want cloud code to just like give up and stop actually working for you, you want codeex to do the work, but you still want to be inside the cloud code ecosystem, maybe it's for a specific feature, we can use the codeex rescue command to do that. So all in all, this is a great plug-in to add to any project you're running with cloud code because it allows you to have a second set of eyes that are anthropics on your code to tell you if what you're doing makes sense.
So before we move on, a quick plug for my cloud code masterass. It is the number one way to go from zero to AI dev, no matter your technical background or lack thereof. We focus on real use cases and it is updated every single week. So, if you want to really master this tool, check it out. It's inside of Chase AI Plus. I'll put a link to that down in the comment. And also, if you're just getting started, I also have the free Chase AI community. There's a link to that in the description.
Now, plugin number two is probably one you've heard a lot about lately, and that is Obsidian in conjunction with the Obsidian skills. So, Obsidian is just a way for us to organize markdown files to organize text files. It's completely free and it's super easy to set up with Claude Code. If you go to obsidian.md, you can download it for free, install it, and all you do is you set one file on your computer as the vault. I call mine the vault. You can call it whatever you want. From there, you just open up Claude Code inside the vault, like I did here, and you're pretty much ready to go. From that point forward, everything Claude Code creates inside of that vault. And if it's a markdown file, will become part of Obsidian. It creates these cool knowledge graphs like this. But what's actually cool about it is it makes it really easy for you to organize stuff. Obsidian sort of becomes a miniature rag system without all the overhead. This is something we walked through in a recent video and I'll link it above where I showed you how to create a vault system with Obsidian and Claude Code that mimics what Carpathy was talking about in a recent viral tweet. This sort of very simple to set up folder system allows us to have a clear pathway where we have Claude Code research things for us and then take that research and turn them into different wiki articles that are simply organized by subfolders. That means we can have a system with hundreds if not thousands of documents that are easy for both Claude Code and us to navigate. And the easiest way to make sure Claude Code is using Obsidian to its full potential is to install the Obsidian skills. This is from the creator, from the CEO of Obsidian himself. And it's just a simple GitHub repo that includes a set of skills that teach Claude Code how to best use it. This sort of setup is perfect if you're using Claude Code in any sort of personal assistant type project or any project that has a large and growing corpus of markdown files.
And speaking of Carpathy, the third tool you want to add to your Claude Code toolbox is Auto Research. So Auto Research is essentially a machine learning algorithm in a box that we can use with Claude Code to improve anything we want. We simply install Auto Research onto our computer using a few lines of code seen here. And then we simply open up Claude Code and say, "Hey, I have some sort of program I'm trying to optimize. I have some sort of skill I'm trying to make better. Let's use auto research to do this." and it's going to run essentially a bunch of machine learning experiments on that program that you're trying to improve all automatically. It will discard any changes that don't improve the program. It will commit any changes that do. And over time, like you see here, you get a better product. And again, this runs all automatically. You don't really have to do anything. So, auto research is perfect if you're doing anything in cloud code where you're trying to optimize something.
Now the fourth thing I want to talk about is this brand new repo and set of skills. It's called awesomedesign.mmd and it is all about front-end design because claude code sucks at it even with some of the bakedin skills like front-end design. It's just not very good. This is a huge step forward. So, design MD takes a lot of its inspiration from Google Stitch, which is what you see here. Google Stitch is an AI tool that's all about front end design. And one of the things Google Stitch does is it creates design files for any website it's creating. And that design MD file, which you see over here on the right, looks something like this. I have it here inside of the notepad, so hopefully it's a little easier to see. It's just a very detailed, very wellthoughtout prompt structure for creating front-end designs, for creating web pages. This isn't something as simple as, hey, don't make it look like AI slop and change how the bento boxes look. This is much more detailed. And that is what awesome.md is based on. What this repo does is it's taken a bunch of popular websites like Claude, like 11 Loves, like Coher, on and on and on, and it's turned their websites into a design. Markdown file like this. What does that mean for you? Well, that means when you're designing a website and you have some sort of inspiration, say from a website like Figma or Notion or Pinterest and you want to use that as a foundation for what you're about to build, well, you just come in here, you click on one of these markdown files, and it essentially gives you a template in that style that you can use for your web page. Everything from buttons to colors to fonts, all that. This is a great jumping off point instead of just praying via prompt that Claude gives you something you like. And to use it is very easy. I would just copy this command, bring it inside of clawed code, and it's going to have all of this ready to go. This repo just came out last week. It's already at 38,000 stars. So, if you're doing anything that requires a web page that you don't want to look like crap, definitely check this out.
Now, for number five, let's talk about some CLI tools and specifically FireCross CLI. and the fire crawl skill. For the most part, every time we talk about a CLI tool, a command line interface, you're going to want a skill as well because the skill teaches cloud code how to use the CLI. So, they're almost always a package deal. Now, fire crawl is perfect if you're doing any sort of web scraping, especially if you're doing web scraping that has a lot of antibbot protections built in because if you just rely on clawed code, standard web search, you're going to run into issues on some of these more protected websites. Firecrawl can get through all of that. Now, the big thing with Firecrawl isn't that it just can go past these B protections, but that when it goes and grabs data from the web, it brings it back in a structured format that's easier for large language models to read. Now, like all these CLI tools, installing this is literally one line of code. And while the main fire crawl product does require you to pay for an API, Firecrawl also is open source. You won't have the proprietary fire crawl engine, which is essentially how they get past the more powerful bots. But if you're just trying to do some basic web scraping and you don't need insane like bot bot detection busters, then you might want to just use the open- source version as well. So you have some options here.
Now moving on to another CLI that's also in the browser automation space, and that is the Playright CLI. Nice thing about this one, totally free. Well, you still pay for tokens, but it's not like the fire crawl web app version that requires us to pay for an API. So the Playright CLI is the latest and greatest version of Playright. It is better than the Playright MCP. So if anyone tells you use the MCP, don't listen to them. The CLI is more effective and it is cheaper. And so the Playright CLI, what it does, it allows us to have Cloud Code do browsers automation on our behalf. Now what do I mean by that? I mean, cloud code using Playright will actually create its own instances of Chrome as if you opened up Chrome and then do things on them like log into websites and test things. Now, there's actually a lot going on with Playright, it has a million different commands, but because you're using inside of Cloud Code, you don't need to know everything that you see here. Claude Code understands what the commands are supposed to do. So once you install the playright CLI, you can just say something like, "Hey Claude Code, I want you to create new instances of Chrome and I want you to test a new website I just created, like test a form submission." And it will create as many Google Chrome tabs essentially as you wanted to. And this is way better than the built-in Claude with Chrome that you've seen running around. So there is a Claude in Chrome Google extension where it will bring up a tab and Claude will do things for you. That isn't great because the cloud code with Chrome is based off of screenshots which are very slow and very expensive. That's not how Playright looks. That's not how Playright looks at the web page and executes things. Playright is actually looking at code under the hood. It's looking at what's called an accessibility tree which makes it way more effective. So if you're doing anything where you need cloud code to actually manipulate web pages on the internet, this is what you want to use.
Now tool number seven is one of my favorites and that is notebook lm-pine. If you've watched my channel at all, you've definitely seen me talk about this one. This CLI tool alongside its skills allows us to essentially hook up Cloud Code to the Notebook LM web app. Because the standard notebook LM doesn't have an API, we can't naturally have Cloud Code talk to it. Yet, by using the CLI tool, we can do exactly that. which means anything you can do normally on notebook LM we can now do inside of cloud code and more because things like batch downloads slide revision full text access and programmatic sharing that's not available on the actual notebook LM web app but it's available through this system we use through the CLI tool and this also includes all the deliverables so things like videos slide decks reports everything you can now do via cloud code and the best part is is it's essentially free. Everything that cla code would be doing for analysis when you put in sources, whether it's YouTube videos or PDFs or whatever, well, that's actually offloaded to Google servers in this case. So, if you're somebody who's struggling with cloud code usage, combining it with something like notebook LM is a godsend because you're using way less tokens. Google's doing all the analysis and then it just hands it to cloud code and you like all these tools, installation is very simple. It's a couple lines of code. And frankly with all these tools when it comes to installation, you can just tell cloud code what you're trying to do and it will do it for you. You can just give it the URL for this GitHub. You can copy and paste the whole page. All those things work. Cloud code is smart enough to understand what commands it needs to run.
Now number eight is a skill and arguably the most important skill you can learn and that is the skill creator skill. Now what makes this skill so much more powerful than the rest isn't that it just creates new skills. It's that it can measure skill performance. So, if you're trying to create a custom skill, you need to know if this new skill actually improves whatever output you're trying to generate versus not having it at all. Before this skill creator skill, we really couldn't do that. Well, at least we couldn't do it easily. But now, by using the skill creator skill, it will actually run benchmarks and run AB tests and give you quantifiable data from which you can start making decisions. And it isn't just with the skill or without the skill. By using the skill, we can test different improvements we're trying to make. So, if I edit a skill because I think it's better, again, it can benchmark it in the same way. And this is huge because skills are arguably the most powerful native feature inside of Claude Code. So, if you ever are going to create your own skill or edit an existing skill, you need to use this. And it's very easy to install because it's an official Claude Code plugin. So, you're just going to go /plugin. You're going to be brought to the marketplace. and then you just search for skill creator skill and install it.
Now number nine is a little niche but it's one of my favorites and that is light rag. This is an open-source graph rag system that is perfect if you're trying to create something that's more robust than obsidian. Now obsidian is obviously amazing but there is a certain scale at which it begins to break down and it is in fact cheaper and quicker to use a proper rag system. Light rag is perfect for this primarily because it's lightweight and it's free. we don't have to go out there and pay for some wildly expensive graph rag system like graph rag itself from Microsoft. So if you're looking for something a little more robust, maybe you're doing a client project or you're dealing with something that is thousands and thousands of documents in size, light rag is perfect for you.
And last but not least is GWS. This is the CLI tool that will allow you to connect cloud code and your entire Google suite. This is super powerful, especially if you're trying to use cloud code as any sort of personal assistant. If you want cloud code to interact with your email, your docs, your calendar, you need this because there's been a lot of hacky things in the past that tried to do it, but this was created by Google developers. It is not an official Google product, but the Google team built this. So, for all intents and purposes, it's one step below having the official Google stamp. Now, the one downside with GWS is the actual setup can be a little laborious and technical because you have to run through the Google Cloud and enable a lot of things. is I will put a link to a video above that can help with that. But the other thing you need to keep in mind is the amount of skills that are available to you if you use GWS. You can see right here all the different skills that are related to using GWS. And that's not necessarily a bad thing. A lot of these skills are essentially tailorade workflows. Things like rescheduling a meeting or organizing your drive folder or scheduling a recurring event. So instead of just hoping Cloud Code can, you know, gobble all the original skills together and figure out how to make the workflow work, well, Google was nice enough or their team at least was nice enough to give us all these recipes just for free. We don't have to think about it. But there's so many you don't want all these loaded at once. So what I suggest you do is point Claude code at this repo and simply say, "Hey, which one of these are actually relevant for us and what we do in our day-to-day?" So if you're someone who's trying to use clawed code as a personal assistant as essentially what OpenClaw is presented as, then GWS is something you want to use.
So if you're new, I hope you were able to take a lot out of this video. If you're someone who's a bit more experienced, I hope at least two or three of those were brand new to you because there are so many tools, so many CLIs, so many plugins that are being added to the Cloud Code ecosystem every single day. It just gets very overwhelming. So, I wanted to try to cut through all the noise and give you what I actually use on a very regular basis. So, as always, let me know in the comments what you thought. Make sure to check out Chase AI Plus if you want to get your hands on that cloud code masterass. and I'll see you around.